mlsanityData to 7 Oct 2026

Breaking the Training Barrier of Billion-Parameter Universal Machine Learning Interatomic Potentials

arXiv 2604.15821 ↗First posted 17 Apr 2026cs.DC cs.LGcited by 0

On the map: Mixture-of-experts → MoE theory and gating functions

Abstract (arXiv)

Universal Machine Learning Interatomic Potentials (uMLIPs), pre-trained on massively diverse datasets encompassing inorganic materials and organic molecules across the entire periodic table, serve as foundational models for quantum-accurate physical simulations. However, uMLIP training requires second-order derivatives, which lack corresponding parallel training frameworks; moreover, scaling to the billion-parameter regime causes explosive growth in computation and communication overhead, making its training a tremendous challenge. We introduce MatRIS-MoE, a billion-parameter Mixture-of-Experts model built upon invariant architecture, and Janus, a pioneering high-dimensional distributed training framework for uMLIPs with hardware-aware optimizations. Deployed across two Exascale supercomputers, our code attains a peak performance of 1.2/1.0 EFLOPS (24%/35.5% of theoretical peak) in single precision at over 90% parallel efficiency, compressing the training of billion-parameter uMLIPs from weeks to hours. This work establishes a new high-water mark for AI-for-Science (AI4S) foundation models at Exascale and provides essential infrastructure for rapid scientific discovery.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in MoE theory and gating functions, one of 12 directions in Mixture-of-experts. Builds on 12 of its 36 references that are in the map. First posted on arXiv 17 Apr 2026.

Builds on

12 of 36 references.

References · 12

  1. 2021E(3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentialscited by 701
  2. 2022MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fieldscited by 467
  3. 2020The Open Catalyst 2020 (OC20) Dataset and Community Challengescited by 246
  4. 2023EquiformerV2: Improved Equivariant Transformer for Scaling to Higher-Degree Representationscited by 201
  5. 2025UMA: A Family of Universal Models for Atomscited by 99
  6. 2024Open Materials 2024 (OMat24) Inorganic Materials Dataset and Modelscited by 50
  7. 2022Towards Training Billion Parameter Graph Neural Networks for Atomic Simulationscited by 25
  8. 2026Pushing the limits of unconstrained machine-learned interatomic potentialscited by 11
  9. 2025MOFSimBench: Evaluating Universal Machine Learning Interatomic Potentials In Metal--Organic Framework Molecular Modelingcited by 7
  10. 2025The Open DAC 2025 Dataset for Sorbent Discovery in Direct Air Capturecited by 7
  11. 2026MatRIS: Toward Reliable and Efficient Pretrained Machine Learning Interatomic Potentialscited by 6
  12. 2026LAER-MoE: Load-Adaptive Expert Re-layout for Efficient Mixture-of-Experts Trainingcited by 5

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Mixture of experts architectures for machine learning interatomic potentials
  2. 2025Automated Machine Learning Pipeline: Large Language Models-Assisted Automated Dataset Generation for Training Machine-Learned Interatomic Potentials
  3. 2026MLIPilot: LLM-Driven Auto-Research for Machine-Learned Interatomic Potentials
  4. 2026Mixture of Experts Framework in Machine Learning Interatomic Potentials for Atomistic Simulations
  5. 2026Lang2MLIP: End-to-End Language-to-Machine Learning Interatomic Potential Development with Autonomous Agentic Workflows
  6. 2026Physics-Aligned Electronic Ground-State Learning Improves Generalization
  7. 2026LLM-Guided Test-Time Discovery of Quantum-Chemical Approximation Algorithms
  8. 2026Equivariant Evidential Deep Learning for Interatomic Potentials